Towards Complex Team Behaviour in Multi-Agent Systems
暂无分享,去创建一个
Team Based Agent Technology (TeBAT) is a programming framework supporting specification of coordinated activity among software agents. It was developed for simulations of land operations in defence, and incorporates both automated path finding in complex environments and distributed control of team movement and operations. As a generic framework, it is applicable to a larger range of domains, including the coordinated operations of autonomous vehicles. From the initial evaluation of existing team approaches, one of the outstanding lessons was that team behaviour cannot be ‘bolted on’ to an agent or multi-agent architecture. Rather, teamwork, and support for teamwork, must be built in from the beginning. The approach taken in TeBAT is based upon the BDI (Belief, Desire, Intention) paradigm, which evolved from early work by Bratman[3] on rational agency. Key to this approach is an emphasis on ‘intentionality’, with teams of agents possessing collaborative intentions. In TeBAT, team movement is characterised as structured flocking behaviours with formation control. Conceptually, each team entity is aware of its place in the formation, and it is aware of any entities around it as well as the terrain and environment in which it operates. The team path finding implements the A* and Dijkstra's search algorithms with modifications: node weightings include team dependent modifiers such as team structure, goals of the team and team knowledge that allow weighting factors to become dynamic. TeBAT further enables dynamic formation and reformation of teams, reasoning over team goal failures at the team level, as well as automatic sharing and aggregation of beliefs betweens teams and sub-teams. Applied to land combat simulations, TeBAT has provided a simplifying infrastructure on top of JACK Intelligent Agents[4]. TeBAT allows team based tactical operations of military doctrine to be captured in an effective way and be played out in simulation scenarios with minimal effort. It avoids the previous laborious construction of complex scripts of detailed entity control, and instead provides a simulation building environment that incorporates team tactics as plug-and-play models. This paper presents the key concepts of the TeBAT modelling framework, and illustrates its application to cases of coordinated activity. We show that although JACK Intelligent AgentsTM provides a full-featured agent development platform, the extension provided by the TeBAT framework brings a significant reduction to the complexity of making agents act as a team. The TeBAT approach to team behaviour is flexible, robust and scalable.
[1] Craig W. Reynolds. Flocks, herds, and schools: a distributed behavioral model , 1998 .
[2] Michael E. Bratman,et al. Intention, Plans, and Practical Reason , 1991 .